public virtual void TestMissFinalizeAndNextStart()
{
// Logger 0: miss finalize(1-3) and start(4)
TestQuorumJournalManagerUnit.FutureThrows(new IOException("injected")).When(spies
[0]).FinalizeLogSegment(Org.Mockito.Mockito.Eq(1L), Org.Mockito.Mockito.Eq(3L));
TestQuorumJournalManagerUnit.FutureThrows(new IOException("injected")).When(spies
[0]).StartLogSegment(Org.Mockito.Mockito.Eq(4L), Org.Mockito.Mockito.Eq(NameNodeLayoutVersion
.CurrentLayoutVersion));
// Logger 1: fail at txn id 4
FailLoggerAtTxn(spies[1], 4L);
QJMTestUtil.WriteSegment(cluster, qjm, 1, 3, true);
EditLogOutputStream stm = qjm.StartLogSegment(4, NameNodeLayoutVersion.CurrentLayoutVersion
);
try
{
QJMTestUtil.WriteTxns(stm, 4, 1);
NUnit.Framework.Assert.Fail("Did not fail to write");
}
catch (QuorumException qe)
{
// Should fail, because logger 1 had an injected fault and
// logger 0 should detect writer out of sync
GenericTestUtils.AssertExceptionContains("Writer out of sync", qe);
}
finally
{
stm.Abort();
qjm.Close();
}
// State:
// Logger 0: 1-3 in-progress (since it missed finalize)
// Logger 1: 1-3 finalized
// Logger 2: 1-3 finalized, 4 in-progress with one txn
// Shut down logger 2 so it doesn't participate in recovery
cluster.GetJournalNode(2).StopAndJoin(0);
qjm = CreateSpyingQJM();
long recovered = QJMTestUtil.RecoverAndReturnLastTxn(qjm);
NUnit.Framework.Assert.AreEqual(3L, recovered);
}
/// <summary>
/// Set up the loggers into the following state:
/// - JN0: edits 1-3 in progress
/// - JN1: edits 1-4 in progress
/// - JN2: edits 1-5 in progress
/// None of the loggers have any associated paxos info.
/// </summary>
/// <exception cref="System.Exception"/>
private void SetupLoggers345()
{
EditLogOutputStream stm = qjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
FailLoggerAtTxn(spies[0], 4);
FailLoggerAtTxn(spies[1], 5);
QJMTestUtil.WriteTxns(stm, 1, 3);
// This should succeed to 2/3 loggers
QJMTestUtil.WriteTxns(stm, 4, 1);
// This should only succeed to 1 logger (index 2). Hence it should
// fail
try
{
QJMTestUtil.WriteTxns(stm, 5, 1);
NUnit.Framework.Assert.Fail("Did not fail to write when only a minority succeeded"
);
}
catch (QuorumException qe)
{
GenericTestUtils.AssertExceptionContains("too many exceptions to achieve quorum size 2/3"
, qe);
}
}
private long WriteSegmentUntilCrash(MiniJournalCluster cluster, QuorumJournalManager
qjm, long txid, int numTxns, Holder <Exception> thrown)
{
long firstTxId = txid;
long lastAcked = txid - 1;
try
{
EditLogOutputStream stm = qjm.StartLogSegment(txid, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (int i = 0; i < numTxns; i++)
{
QJMTestUtil.WriteTxns(stm, txid++, 1);
lastAcked++;
}
stm.Close();
qjm.FinalizeLogSegment(firstTxId, lastAcked);
}
catch (Exception t)
{
thrown.held = t;
}
return(lastAcked);
}
/// <summary>
/// Test the case where, at the beginning of a segment, transactions
/// have been written to one JN but not others.
/// </summary>
/// <exception cref="System.Exception"/>
public virtual void DoTestOutOfSyncAtBeginningOfSegment(int nodeWithOneTxn)
{
int nodeWithEmptySegment = (nodeWithOneTxn + 1) % 3;
int nodeMissingSegment = (nodeWithOneTxn + 2) % 3;
QJMTestUtil.WriteSegment(cluster, qjm, 1, 3, true);
WaitForAllPendingCalls(qjm.GetLoggerSetForTests());
cluster.GetJournalNode(nodeMissingSegment).StopAndJoin(0);
// Open segment on 2/3 nodes
EditLogOutputStream stm = qjm.StartLogSegment(4, NameNodeLayoutVersion.CurrentLayoutVersion
);
try
{
WaitForAllPendingCalls(qjm.GetLoggerSetForTests());
// Write transactions to only 1/3 nodes
FailLoggerAtTxn(spies[nodeWithEmptySegment], 4);
try
{
QJMTestUtil.WriteTxns(stm, 4, 1);
NUnit.Framework.Assert.Fail("Did not fail even though 2/3 failed");
}
catch (QuorumException qe)
{
GenericTestUtils.AssertExceptionContains("mock failure", qe);
}
}
finally
{
stm.Abort();
}
// Bring back the down JN.
cluster.RestartJournalNode(nodeMissingSegment);
// Make a new QJM. At this point, the state is as follows:
// A: nodeWithEmptySegment: 1-3 finalized, 4_inprogress (empty)
// B: nodeWithOneTxn: 1-3 finalized, 4_inprogress (1 txn)
// C: nodeMissingSegment: 1-3 finalized
GenericTestUtils.AssertGlobEquals(cluster.GetCurrentDir(nodeWithEmptySegment, QJMTestUtil
.Jid), "edits_.*", NNStorage.GetFinalizedEditsFileName(1, 3), NNStorage.GetInProgressEditsFileName
(4));
GenericTestUtils.AssertGlobEquals(cluster.GetCurrentDir(nodeWithOneTxn, QJMTestUtil
.Jid), "edits_.*", NNStorage.GetFinalizedEditsFileName(1, 3), NNStorage.GetInProgressEditsFileName
(4));
GenericTestUtils.AssertGlobEquals(cluster.GetCurrentDir(nodeMissingSegment, QJMTestUtil
.Jid), "edits_.*", NNStorage.GetFinalizedEditsFileName(1, 3));
// Stop one of the nodes. Since we run this test three
// times, rotating the roles of the nodes, we'll test
// all the permutations.
cluster.GetJournalNode(2).StopAndJoin(0);
qjm = CreateSpyingQJM();
qjm.RecoverUnfinalizedSegments();
if (nodeWithOneTxn == 0 || nodeWithOneTxn == 1)
{
// If the node that had the transaction committed was one of the nodes
// that responded during recovery, then we should have recovered txid
// 4.
CheckRecovery(cluster, 4, 4);
QJMTestUtil.WriteSegment(cluster, qjm, 5, 3, true);
}
else
{
// Otherwise, we should have recovered only 1-3 and should be able to
// start a segment at 4.
CheckRecovery(cluster, 1, 3);
QJMTestUtil.WriteSegment(cluster, qjm, 4, 3, true);
}
}
